A Novel Image Classification Method Based on Multi-layer Dictionary Learning
Dandan Zhao, Minhan Yi, Jiaxin Guo, Hongpeng Yin · 2021 CAA Symposium on Fault Detection, Supervision, and Safety for Technical Processes (SAFEPROCESS) · 2021
Deep learning is widely used in image classification due to the huge potential of multi-layer training. Combining deep learning with traditional dictionary learning has become a hot topic. However, the existing methods update dictionary learning and classifier as two independent modules, which limits the classification performance. In this paper, a novel image classification method based on multi-layer dictionary learning is proposed to meet this challenge. First, multi-layer training is introduced to extract the multi-layer nonlinear features of image data. Then, the discriminant label constraint term is designed to optimize the dictionary and classifier synchronously. The proposed method improves classification accuracy via enhancing the coding coefficient features and screening the beneficial classification feature. Finally, we examine the performance of our model on two real-world image data sets. The correctness and effectiveness of the proposed model are verified through convergence, accuracy and parameter sensitivity.